Horowitz argues AI apps rely on proprietary data for defensibility
Ben Horowitz contends that AI applications are sophisticated systems, not simple wrappers, citing proprietary data as a key differentiator. He predicts long-term defensibility will come from customer retention rather than network effects.
Ben Horowitz argues that the characterization of AI applications as mere "GPT wrappers" is incorrect, comparing the current landscape to the early days of relational databases where application layers like SAP and Salesforce proved durable despite the underlying technology. He notes that modern AI apps, such as Cursor, utilize sophisticated stacks with multiple models, citing 14 models for Cursor, built on top of foundation models. These layers accumulate proprietary data and a deep understanding of specific user bases, such as high-end programmers, creating differentiation that competitors cannot easily replicate.
Horowitz emphasizes that in the current "land grab" phase, the primary feature is functionality, but long-term defensibility comes from possessing the customer and having a dramatically better product. He observes that switching costs are low and network effects are not yet fully established in the AI space, meaning companies must rely on product superiority and data accumulation to maintain their position. This perspective challenges the notion that foundation models alone determine competitive advantage, instead highlighting the value of the application layer in capturing and leveraging user-specific insights.